Message-Id: <1465327445-56484-2-git-send-email-Waiman.Long@hpe.com>
Date:	Tue,  7 Jun 2016 15:24:02 -0400
From:	Waiman Long <Waiman.Long@....com>
To:	Alexander Viro <viro@...iv.linux.org.uk>, Jan Kara <jack@...e.com>,
	Jeff Layton <jlayton@...chiereds.net>,
	"J. Bruce Fields" <bfields@...ldses.org>,
	Tejun Heo <tj@...nel.org>,
	Christoph Lameter <cl@...ux-foundation.org>
Cc:	linux-fsdevel@...r.kernel.org, linux-kernel@...r.kernel.org,
	Ingo Molnar <mingo@...hat.com>,
	Peter Zijlstra <peterz@...radead.org>,
	Andi Kleen <andi@...stfloor.org>,
	Dave Chinner <dchinner@...hat.com>,
	Boqun Feng <boqun.feng@...il.com>,
	Scott J Norton <scott.norton@....com>,
	Douglas Hatch <doug.hatch@....com>,
	Waiman Long <Waiman.Long@....com>
Subject: [RESEND PATCH v7 1/4] lib/percpu-list: Per-cpu list with associated per-cpu locks

Linked list is used everywhere in the Linux kernel. However, if many
threads are trying to add or delete entries into the same linked list,
it can create a performance bottleneck.

This patch introduces a new per-cpu list subystem with associated
per-cpu locks for protecting each of the lists individually. This
allows list entries insertion and deletion operations to happen in
parallel instead of being serialized with a global list and lock.

List entry insertion is strictly per cpu. List deletion, however, can
happen in a cpu other than the one that did the insertion. So we still
need lock to protect the list. Because of that, there may still be
a small amount of contention when deletion is being done.

A new header file include/linux/percpu-list.h will be added with the
associated pcpu_list_head and pcpu_list_node structures. The following
functions are provided to manage the per-cpu list:

 1. int init_pcpu_list_head(struct pcpu_list_head **ppcpu_head)
 2. void pcpu_list_add(struct pcpu_list_node *node,
		       struct pcpu_list_head *head)
 3. void pcpu_list_del(struct pcpu_list *node)

Iteration of all the list entries within a group of per-cpu
lists is done by calling either the pcpu_list_iterate() or
pcpu_list_iterate_safe() functions in a while loop. They correspond
to the list_for_each_entry() and list_for_each_entry_safe() macros
respectively. The iteration states are keep in a pcpu_list_state
structure that is passed to the iteration functions.

Signed-off-by: Waiman Long <Waiman.Long@....com>
---
 include/linux/percpu-list.h |  231 +++++++++++++++++++++++++++++++++++++++++++
 lib/Makefile                |    2 +-
 lib/percpu-list.c           |  100 +++++++++++++++++++
 3 files changed, 332 insertions(+), 1 deletions(-)
 create mode 100644 include/linux/percpu-list.h
 create mode 100644 lib/percpu-list.c

diff --git a/include/linux/percpu-list.h b/include/linux/percpu-list.h
new file mode 100644
index 0000000..ce8238a
--- /dev/null
+++ b/include/linux/percpu-list.h
@@ -0,0 +1,231 @@
+/*
+ * Per-cpu list
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * (C) Copyright 2016 Hewlett-Packard Enterprise Development LP
+ *
+ * Authors: Waiman Long <waiman.long@....com>
+ */
+#ifndef __LINUX_PERCPU_LIST_H
+#define __LINUX_PERCPU_LIST_H
+
+#include <linux/spinlock.h>
+#include <linux/list.h>
+#include <linux/percpu.h>
+
+/*
+ * include/linux/percpu-list.h
+ *
+ * A per-cpu list protected by a per-cpu spinlock.
+ *
+ * The pcpu_list_head structure contains the spinlock, the other
+ * pcpu_list_node structures only contains a pointer to the spinlock in
+ * pcpu_list_head.
+ */
+struct pcpu_list_head {
+	struct list_head list;
+	spinlock_t lock;
+};
+
+#define PCPU_LIST_HEAD_INIT(name)				\
+	{							\
+		.list.prev = &name.list,			\
+		.list.next = &name.list,			\
+		.list.lock = __SPIN_LOCK_UNLOCKED(name),	\
+	}
+
+/*
+ * Per-cpu list iteration state
+ */
+struct pcpu_list_state {
+	int			 cpu;
+	spinlock_t		*lock;
+	struct list_head	*head;	/* List head of current per-cpu list */
+	struct pcpu_list_node	*curr;
+	struct pcpu_list_node	*next;
+};
+
+#define PCPU_LIST_STATE_INIT()			\
+	{					\
+		.cpu  = -1,			\
+		.lock = NULL,			\
+		.head = NULL,			\
+		.curr = NULL,			\
+		.next = NULL,			\
+	}
+
+#define DEFINE_PCPU_LIST_STATE(s)		\
+	struct pcpu_list_state s = PCPU_LIST_STATE_INIT()
+
+static inline void init_pcpu_list_state(struct pcpu_list_state *state)
+{
+	state->cpu  = -1;
+	state->lock = NULL;
+	state->head = NULL;
+	state->curr = NULL;
+	state->next = NULL;
+}
+
+#ifdef CONFIG_DEBUG_SPINLOCK
+#define PERCPU_LIST_WARN_ON(x)	WARN_ON(x)
+#else
+#define PERCPU_LIST_WARN_ON(x)
+#endif
+
+/*
+ * Next per-cpu list entry
+ */
+#define pcpu_list_next_entry(pos, member) list_next_entry(pos, member.list)
+
+/*
+ * Per-cpu node data structure
+ */
+struct pcpu_list_node {
+	struct list_head list;
+	spinlock_t *lockptr;
+};
+
+#define PCPU_LIST_NODE_INIT(name)		\
+	{					\
+		.list.prev = &name.list,	\
+		.list.next = &name.list,	\
+		.list.lockptr = NULL		\
+	}
+
+static inline void init_pcpu_list_node(struct pcpu_list_node *node)
+{
+	INIT_LIST_HEAD(&node->list);
+	node->lockptr = NULL;
+}
+
+static inline void free_pcpu_list_head(struct pcpu_list_head **ppcpu_head)
+{
+	free_percpu(*ppcpu_head);
+	*ppcpu_head = NULL;
+}
+
+/*
+ * Check if all the per-cpu lists are empty
+ */
+static inline bool pcpu_list_empty(struct pcpu_list_head *pcpu_head)
+{
+	int cpu;
+
+	for_each_possible_cpu(cpu)
+		if (!list_empty(&per_cpu_ptr(pcpu_head, cpu)->list))
+			return false;
+	return true;
+}
+
+/*
+ * Helper function to find the first entry of the next per-cpu list
+ * It works somewhat like for_each_possible_cpu(cpu).
+ *
+ * Return: true if the entry is found, false if all the lists exhausted
+ */
+static __always_inline bool
+__pcpu_list_next_cpu(struct pcpu_list_head *head, struct pcpu_list_state *state)
+{
+	if (state->lock)
+		spin_unlock(state->lock);
+next_cpu:
+	/*
+	 * for_each_possible_cpu(cpu)
+	 */
+	state->cpu = cpumask_next(state->cpu, cpu_possible_mask);
+	if (state->cpu >= nr_cpu_ids)
+		return false;	/* All the per-cpu lists iterated */
+
+	state->head = &per_cpu_ptr(head, state->cpu)->list;
+	if (list_empty(state->head))
+		goto next_cpu;
+
+	state->lock = &per_cpu_ptr(head, state->cpu)->lock;
+	spin_lock(state->lock);
+	/*
+	 * There is a slight chance that the list may become empty just
+	 * before the lock is acquired. So an additional check is
+	 * needed to make sure that state->curr points to a valid entry.
+	 */
+	if (list_empty(state->head)) {
+		spin_unlock(state->lock);
+		goto next_cpu;
+	}
+	state->curr = list_entry(state->head->next,
+				 struct pcpu_list_node, list);
+	return true;
+}
+
+/*
+ * Iterate to the next entry of the group of per-cpu lists
+ *
+ * Return: true if the next entry is found, false if all the entries iterated
+ */
+static inline bool pcpu_list_iterate(struct pcpu_list_head *head,
+				     struct pcpu_list_state *state)
+{
+	/*
+	 * Find next entry
+	 */
+	if (state->curr)
+		state->curr = list_next_entry(state->curr, list);
+
+	if (!state->curr || (&state->curr->list == state->head)) {
+		/*
+		 * The current per-cpu list has been exhausted, try the next
+		 * per-cpu list.
+		 */
+		if (!__pcpu_list_next_cpu(head, state))
+			return false;
+	}
+
+	PERCPU_LIST_WARN_ON(state->curr->lockptr != state->lock);
+	return true;	/* Continue the iteration */
+}
+
+/*
+ * Iterate to the next entry of the group of per-cpu lists and safe
+ * against removal of list_entry
+ *
+ * Return: true if the next entry is found, false if all the entries iterated
+ */
+static inline bool pcpu_list_iterate_safe(struct pcpu_list_head *head,
+					  struct pcpu_list_state *state)
+{
+	/*
+	 * Find next entry
+	 */
+	if (state->curr) {
+		state->curr = state->next;
+		state->next = list_next_entry(state->next, list);
+	}
+
+	if (!state->curr || (&state->curr->list == state->head)) {
+		/*
+		 * The current per-cpu list has been exhausted, try the next
+		 * per-cpu list.
+		 */
+		if (!__pcpu_list_next_cpu(head, state))
+			return false;
+		state->next = list_next_entry(state->curr, list);
+	}
+
+	PERCPU_LIST_WARN_ON(state->curr->lockptr != state->lock);
+	return true;	/* Continue the iteration */
+}
+
+extern void pcpu_list_add(struct pcpu_list_node *node,
+			  struct pcpu_list_head *head);
+extern void pcpu_list_del(struct pcpu_list_node *node);
+extern int  init_pcpu_list_head(struct pcpu_list_head **ppcpu_head);
+
+#endif /* __LINUX_PERCPU_LIST_H */
diff --git a/lib/Makefile b/lib/Makefile
index a65e9a8..387ed2b 100644
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -40,7 +40,7 @@ obj-y += bcd.o div64.o sort.o parser.o halfmd4.o debug_locks.o random32.o \
 	 gcd.o lcm.o list_sort.o uuid.o flex_array.o iov_iter.o clz_ctz.o \
 	 bsearch.o find_bit.o llist.o memweight.o kfifo.o \
 	 percpu-refcount.o percpu_ida.o rhashtable.o reciprocal_div.o \
-	 once.o
+	 once.o percpu-list.o
 obj-y += string_helpers.o
 obj-$(CONFIG_TEST_STRING_HELPERS) += test-string_helpers.o
 obj-y += hexdump.o
diff --git a/lib/percpu-list.c b/lib/percpu-list.c
new file mode 100644
index 0000000..8a96001
--- /dev/null
+++ b/lib/percpu-list.c
@@ -0,0 +1,100 @@
+/*
+ * Per-cpu list
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * (C) Copyright 2016 Hewlett-Packard Enterprise Development LP
+ *
+ * Authors: Waiman Long <waiman.long@....com>
+ */
+#include <linux/percpu-list.h>
+#include <linux/lockdep.h>
+
+/*
+ * The per-cpu list lock needs its own class to avoid warning and stack
+ * trace when lockdep is enabled.
+ */
+static struct lock_class_key percpu_list_key;
+
+/*
+ * Initialize the per-cpu list head
+ */
+int init_pcpu_list_head(struct pcpu_list_head **ppcpu_head)
+{
+	struct pcpu_list_head *pcpu_head = alloc_percpu(struct pcpu_list_head);
+	int cpu;
+
+	if (!pcpu_head)
+		return -ENOMEM;
+
+	for_each_possible_cpu(cpu) {
+		struct pcpu_list_head *head = per_cpu_ptr(pcpu_head, cpu);
+
+		INIT_LIST_HEAD(&head->list);
+		head->lock = __SPIN_LOCK_UNLOCKED(&head->lock);
+		lockdep_set_class(&head->lock, &percpu_list_key);
+	}
+
+	*ppcpu_head = pcpu_head;
+	return 0;
+}
+
+/*
+ * List selection is based on the CPU being used when the pcpu_list_add()
+ * function is called. However, deletion may be done by a different CPU.
+ * So we still need to use a lock to protect the content of the list.
+ */
+void pcpu_list_add(struct pcpu_list_node *node, struct pcpu_list_head *head)
+{
+	struct pcpu_list_head *myhead;
+
+	/*
+	 * Disable preemption to make sure that CPU won't gets changed.
+	 */
+	preempt_disable();
+	myhead = this_cpu_ptr(head);
+	spin_lock(&myhead->lock);
+	node->lockptr = &myhead->lock;
+	list_add(&node->list, &myhead->list);
+	spin_unlock(&myhead->lock);
+	preempt_enable();
+}
+
+/*
+ * Delete a node from a percpu list
+ *
+ * We need to check the lock pointer again after taking the lock to guard
+ * against concurrent delete of the same node. If the lock pointer changes
+ * (becomes NULL or to a different one), we assume that the deletion was done
+ * elsewhere.
+ */
+void pcpu_list_del(struct pcpu_list_node *node)
+{
+	spinlock_t *lock = READ_ONCE(node->lockptr);
+
+	if (unlikely(!lock)) {
+		WARN(1, "pcpu_list_del: node 0x%lx has no associated lock\n",
+			(unsigned long)node);
+		return;
+	}
+
+	spin_lock(lock);
+	if (likely(lock == node->lockptr)) {
+		list_del_init(&node->list);
+		node->lockptr = NULL;
+	} else {
+		/*
+		 * This path should never be executed.
+		 */
+		WARN_ON(1);
+	}
+	spin_unlock(lock);
+}
-- 
1.7.1

Powered by blists - more mailing lists